Docs Coverage

Docs coverage

A skill for auditing whether a design system's documentation surface keeps pace with its components. It measures the code (the source of truth for what exists) against each documentation surface and reports three things: coverage gaps (components with no documentation), staleness (documentation that predates the component's last code change), and orphaned docs (pages for components that no longer exist). Produces a severity-rated finding table with per-signal confidence labelling.

Before you begin: verify references

Confirm that every path in this skill's frontmatter references: exists relative to this SKILL.md. If any is missing, stop: the install is incomplete, usually because a flattening installer (for example npx skills install) dropped the repo-root knowledge-notes/ directory. Tell the user to reinstall by a method in 1-INSTALL.md and run verify-install.sh from the install root. Proceed without the references only if the user explicitly says to, and then say in the output that it was produced without the pack's reference material.

Context

Code is the source of truth for what components exist; the documentation surface is measured against it. A component in code with no docs is a coverage gap; a page for a deleted component is an orphan; a page older than the component's last change is a staleness risk.

This skill is built to work with no integration at all: a components directory plus a Storybook build plus git history answer coverage and staleness for most teams. Hosted platforms (Zeroheight, Supernova, custom docs sites) are optional layers that light up when configured — never prerequisites. The audit never blocks on an integration; it logs what is unavailable and proceeds with what it can reach.

The hard part is trust. Coverage is a join between two lists — components in code and entries in the doc surface — and the join is only as reliable as the key that links them. Every coverage finding carries a confidence tier so a fuzzy guess is never presented as a fact. See documentation-coverage.md for the full model.

---

Configuration

If .ds-ops-config.yml exists, follow the configuration-and-recurring knowledge note (../../knowledge-notes/configuration-and-recurring.md) for loading, integration fallbacks and recurring runs. This skill reads:

• system.framework — affects how component files are discovered (e.g. .tsx / .vue / .twig)
• severity.* — severity-rating overrides
• integrations.storybook.static_path — local Storybook build directory (e.g. storybook-static); the preferred source
• integrations.storybook.url — published Storybook URL for pulling /index.json when no local build exists
• integrations.documentation — optional hosted platform: platform, url, api_key_env, plus styleguide_id (Zeroheight) or design_system_id (Supernova)
• integrations.github — change history when the audit runs outside a local clone
• docs_coverage.staleness_threshold_days — grace window before a doc is flagged stale (default 90)
• recurring.* — trend comparison (see Recurring workflow)

Auto-pull integrations

Storybook — the primary surface (integrations.storybook.enabled: true or a local build):

• Prefer a local static build: read the index from integrations.storybook.static_path (default storybook-static/index.json). No server, no auth.
• If only a URL is configured, fetch <url>/index.json.
• No build and no URL: read the CSF files directly. *.stories.* grouped by their title (or the default export's component) give rung 1; a tags: ['autodocs'] entry or a sibling MDX file gives a docs page. Say under Scope that the inventory came from story source, not an index, and that componentPath joins weren't available.
• Branch on the top-level v field, which tracks the Storybook version: v: 3 is the SB 6 stories.json (entries under the stories key); v: 4 and v: 5 are the SB 7+ index.json (entries under entries), and v: 5 (SB 8.1+) adds componentPath. componentPath is opt-in and not guaranteed even on recent Storybook — use it for the Tier A join when present, and fall back to the Tier B name join whenever it is absent, regardless of v. A v: 3 stories.json has no type: 'docs' entries at all, so it can't show rung 2: mark rung 2 unknown for v3, unless you read the MDX files or each story's parameters.docs directly.
• The official Storybook MCP (@storybook/addon-mcp) is optional — at the time of writing it needs a running server and is React-only/experimental. Use it only if the tools are already available; never make it a dependency.

Documentation platform — optional layers (integrations.documentation.enabled: true):

• zeroheight: use the REST API (Enterprise) — GET /styleguides/{id}/pages and GET /pages/{id} give the documented-page set and updated_at per page (high-confidence staleness). There is no components endpoint, so reconstruct the coverage diff by name-matching pages against the code/Storybook inventory (Tier C).
• supernova: use the MCP "Relay" or @supernovaio/sdk — get_design_system_component_list vs get_documentation_page_list gives a coverage diff (heuristic link, Tier B/C). Per-page staleness timestamps are not reliably exposed — mark staleness unknown unless a page timestamp is actually present.
• custom: crawl the sitemap or rendered HTML for page titles; name-match only (Tier C).

Step 0: Build the component inventory (the source of truth)

Establish what components exist before looking at any documentation.

1. If a codebase-index output exists (.ai/index/), use its component list — it is already resolved and classified.
2. Otherwise, if the package has a public barrel (src/index.ts, or each package's entry point in a monorepo), use its component exports as the inventory — that's what consumers see and what needs docs. Internal building blocks that aren't exported don't need public docs; count them separately if at all.
3. Only if there's no barrel, glob the components directory. Common roots: src/components/**, packages/*/src/**, lib/components/**, app/components/**. Treat each component file/symbol as a candidate (e.g. Button.tsx, not Button.test.tsx, Button.stories.tsx, or index.ts barrels), and say the inventory may include internal components.
4. Record, per component: name, resolved file path (repo-relative, forward-slash), and category if a classification is available.

Produce a brief inventory line before continuing: N components found across M directories. If discovery finds nothing, ask the user where components live.

Step 1: Build the documentation inventory

For each available surface, list what is documented.

• Storybook: parse index.json. Group entries by title. For each component, record: has a type: 'story' entry (rung 1, exists); has a docs page — autodocs or MDX via tags (unknown for a v3 stories.json, see above); and the resolved componentPath and importPath. A docs page alone doesn't make a component described: the autodocs tag generates a page even when no prop has a description. Rung 2 needs the page and descriptions on most props, from argTypes in the CSF file or JSDoc on the props interface. Record the prop-description ratio per component (described props of total) and report it; a page with 0 of 12 props described is rung 1 with a docs page, not rung 2. Also note play presence (interaction tests) where the CSF is read.
• Figma component descriptions (Console MCP figma_get_component, or the official MCP's get_design_context on the library node): a documentation surface for designers. Record which components have a non-empty description and report it as its own column, not merged into the Storybook rungs.
• Hosted platform (if configured): list documented pages and, where exposed, their updated_at/last-modified timestamp.
• Usage guidance (rung 3): if the system documents usage separately (a usage-guidelines output, MDX "When to use" sections, a Zeroheight guideline page), record which components reach rung 3.

Step 2: Join the inventories

Join code components to documentation entries using the Tier A / B / C reliability hierarchy in documentation-coverage.md (file path, then symbol name, then fuzzy title match), and record the tier on every match. Never state a Tier C result as fact.

Before reporting anything as undocumented, run a positive control: confirm the join matches a component you can see is documented (open its story or page). If the join misses it, the key doesn't fit this codebase — fix it before reporting gaps.

Step 3: Coverage gap analysis

From the join, produce:

• Undocumented components — in code, no entry on any surface. Report each with the join tier that found (or failed to find) it. Group by the documentation rung they fall short of.
• Rung distribution — how many components reach exists / described / guided. A high story count with few described/guided components is itself a finding.
• Orphaned docs — pages/stories whose component no longer exists in code. These point the opposite direction and are usually quick removals. Before calling a page orphaned, search the codebase for the name it documents (including renamed or re-exported symbols) — a Tier B/C "orphan" is often just a title that drifted from the component name.